Saliency & Biological Vision

The paper frames saliency detection as identifying regions that attract greater visual attention than other parts of an image.

“Such image regions are said to be salient since it is assumed that they attract greater attention by the visual system than other parts of the image.”◌ not checked against the paper’s text as it now stands

The paper focuses on bottom-up saliency mechanisms as a means of selecting attentional regions, distinguishing them from task-dependent top-down attention.

“Generally speaking, there are two different processes that influence visual saliency; one is top-down and depends on the task at hand and the other is bottom-up, which is driven by the input image. The focus of the paper is bottom-up saliency for selecting attentional regions.”◌ not checked against the paper’s text as it now stands

The paper demonstrates that its proposed saliency model can predict where people focus their attention, validated against eye-tracking fixation data.

“In addition, we show that it is able to predict salient regions on which people focus their attention.”◌ not checked against the paper’s text as it now stands

The paper connects its non-saliency suppression approach to biological attention theory, modeling how visual stimuli compete for attention in the visual cortex and are inhibited to allow salient objects to stand out.

“Research has suggested that objects viewed by the human visual system are thought to compete with each other to selectively focus our attention on a subset […]. Objects that appear in the visual field will influence how they are viewed by suppressing each other. Consequently, many are inhibited, while those that are not, will ultimately predominate in the visual cortex to provide a focus of attention.”◌ not checked against the paper’s text as it now stands

The paper explicitly identifies visual attention as the human capacity that computational saliency models are attempting to model.

“VISUAL attention, the astonishing capability of human visual system to selectively process only the salient visual stimuli in details, has been investigated by multiple disciplines such as cognitive psychology, neuroscience, and computer vision […].”◌ not checked against the paper’s text as it now stands

The paper lists visual attention as one of the keywords describing the concepts covered in salient object detection benchmarking.

“Salient object detection, saliency, explicit saliency, visual attention, regions of interest, objectness, segmentation, interestingness, importance, eye movements”◌ not checked against the paper’s text as it now stands

The paper traces computational saliency and attention models back to cognitive attention theories and early attention algorithms as foundational background.

“Following cognitive theories (e.g., feature integration theory (FIT) […], guided search model […]) and early attention models (e.g., Koch and Ullman […] and Itti et al. […]), hundreds of computational saliency models have been proposed to detect salient visual subsets from images and videos.”◌ not checked against the paper’s text as it now stands

The paper draws an analogy between detection proposals and interest point detectors as mechanisms for focusing attention to reduce computation.

“Interest points allow for focusing attention to the most salient and distinctive locations in an image, greatly reducing computation for subsequent tasks such as classification, retrieval, matching, and detection.”◌ not checked against the paper’s text as it now stands

The paper investigates whether attention-focusing proposal mechanisms improve detection accuracy or are merely necessary for computational efficiency.

“We aim to understand if detection proposals improve detection accuracy or if their use is strictly necessary for computational reasons.”◌ not checked against the paper’s text as it now stands

The paper notes a historical shift where dense feature extraction replaced sparse interest-point-based attention mechanisms in general vision tasks, contrasting with the trend in detection.

“It is worthwhile noting that interest points were dominant when computing feature descriptors densely was prohibitive. However, with improved algorithmic efficiency and increased computational power, it is now standard practice to use dense feature extraction […]. The opposite trend has occurred in object detection, where the dense sliding window framework has been overtaken by use of proposals.”

The paper distinguishes its salient object detection task from visual attention modeling, which is a separate line of saliency research focused on predicting human fixations.

“Saliency models have been built for visual attention modeling […] and salient object detection […]. The former task aims to predict human fixation locations on natural images, while the proposed method aims to compute the pixel-wise saliency values for capturing the regions of the salient objects.”◌ not checked against the paper’s text as it now stands

The paper demonstrates that its learned saliency values reflect known human attention biases toward red bright colors.

“Specifically, the saliency value of the red flower in Figure […](a) is higher than the yellow one, which is consistent with human perception (paying more attention to the objects with red bright colors […]).”◌ not checked against the paper’s text as it now stands

The paper explicitly links its qualitative saliency examples to human attention tendencies toward red objects, image center, contrast, and semantic objects.

“Given an image, people tend to pay their attention to red objects […] (a), the image center […] (b), regions with high contrast in texture and shape […] (c), and semantic objects […] (d).”

The paper categorizes saliency detection research into visual attention prediction and salient object detection as two distinct types of methods.

“saliency models in general can be categorized into visual attention prediction or salient object detection.”◌ not checked against the paper’s text as it now stands

The paper explains that visual attention prediction methods aim to predict where a human observer would fixate in a scene.

“The former methods […] try to predict scene locations where a human observer may fixate.”◌ not checked against the paper’s text as it now stands

The paper reviews deep learning models used for attention prediction in dynamic scenes as part of its related work survey.

“including action recognition […], object segmentation […], object tracking […], attention prediction […] and semantic segmentation […], and explore their architectures and training schemes.”◌ not checked against the paper’s text as it now stands

The paper notes that other works have adapted two-stream convolutional network architectures for dynamic attention prediction tasks.

The paper conducts a comprehensive survey categorizing attention mechanisms in computer vision according to the dimension they operate on, such as channel, spatial, temporal, and branch attention.

“we provide a {comprehensive} review of various attention mechanisms in computer vision and categorize them according to %different applied {the} dimensions {they operating on,} such as channel attention, spatial attention {,} %and temporal attention {and branch attention}.”◌ not checked against the paper’s text as it now stands

The paper defines attention mechanism in computer vision as a dynamic selection process realized by adaptively weighting features according to the input.

“In [JPT: the] vision system, the attention mechanism can be treated as a dynamic selection process that is realized by adaptive weighting features according to the change of input.”◌ not checked against the paper’s text as it now stands

The paper traces the historical development of attention in computer vision through four stages: RNN-based attention, explicit region prediction, implicit channel attention, and self-attention.

“The whole progress can be roughly divided into four parts. The first part begins from RAM […], which is the pioneer that combines deep neural networks with attention mechanisms.”